Skip to content

[Bug][QDP] Align Iris GPU training path with sibling pipelines (real dtype + leaf weights) #1412

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ryankert01 merged 2 commits into from
Jul 10, 2026
+11 −6
Merged

[Bug][QDP] Align Iris GPU training path with sibling pipelines (real dtype + leaf weights) #1412

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
dd8bfbf
[Bug][QDP] Use real dtype for weights/labels in Iris GPU training path
ryankert01 Jun 26, 2026
256e70d
[Bug][QDP] Fix non-leaf weight tensor in Iris GPU training path
ryankert01 Jun 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 11 additions & 6 deletions qdp/qdp-python/benchmark/encoding_benchmarks/qdp_pipeline/iris_amplitude.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -322,9 +322,12 @@ def _run_training_gpu(
) -> dict[str, Any]:
"""GPU path: lightning.gpu + PyTorch interface, data stays on GPU. Optional early stop every 100 steps."""
device = encoded_train.device
dtype = encoded_train.dtype
Y_train_t = torch.tensor(Y_train, dtype=dtype, device=device)
Y_test_t = torch.tensor(Y_test, dtype=dtype, device=device)
# Encoded data may be complex (from QDP); use real dtype for weights, bias and labels.
real_dtype = (
torch.float64 if encoded_train.dtype == torch.complex128 else torch.float32
)
Y_train_t = torch.tensor(Y_train, dtype=real_dtype, device=device)
Y_test_t = torch.tensor(Y_test, dtype=real_dtype, device=device)

@qml.qnode(dev_qml, interface="torch", diff_method="adjoint")
def circuit(weights, state_vector):
Expand All @@ -341,10 +344,12 @@ def cost(weights, bias, X_batch, Y_batch):
return torch.mean((Y_batch - preds) ** 2)

torch.manual_seed(seed)
weights = 0.01 * torch.randn(
num_layers, NUM_QUBITS, 3, device=device, dtype=dtype, requires_grad=True
weights = (
(0.01 * torch.randn(num_layers, NUM_QUBITS, 3, device=device, dtype=real_dtype))
.detach()
.requires_grad_(True)
)
bias = torch.tensor(0.0, device=device, dtype=dtype, requires_grad=True)
bias = torch.tensor(0.0, device=device, dtype=real_dtype, requires_grad=True)
opt = torch.optim.SGD([weights, bias], lr=lr, momentum=0.9, nesterov=True)

t0 = time.perf_counter()
Expand Down
Loading